One-loop QED vertex function and the anomalous magnetic moment

Intuition Beginner

A free electron, treated by the Dirac equation, carries a magnetic moment $\mu =(g/2)(e\hslash /2mc)$ with $g=2$. The number $2$ falls out of the non-relativistic limit of the Dirac equation as cleanly as $E=m{c}^2$ falls out of relativistic kinematics; it was Dirac's signature triumph in 1928 and an unmistakable hint that the Dirac equation was on the right track.

Experiments in 1947 by Foley and Kusch then showed that $g$ is not exactly $2$. The measured value sat a tenth of a percent higher, near $g=2.00238$. The excess $a_e=(g-2)/2\approx 1.19\times 10^{-3}$ became known as the anomalous magnetic moment. Within a year Julian Schwinger, working in his New York apartment, computed it from quantum electrodynamics and reproduced the experiment.

The mechanism is that an electron is never alone. Even sitting still, it is constantly emitting and reabsorbing virtual photons. When you probe the electron with an external magnetic field, you do not couple to the bare electron but to the electron-plus-cloud, and the cloud changes the effective coupling by a small amount. The leading correction comes from a single virtual photon: in Schwinger's calculation the result is $a_e=\alpha /(2\pi )$, where $\alpha =e^2/(4\pi {ϵ}_0\hslash c)\approx 1/137$ is the fine-structure constant. Numerically this gives $a_e\approx 0.001162$, agreeing with Foley-Kusch and with the seventy years of ever-more-precise measurements since.

That number — $\alpha /(2\pi )\approx 0.00116$ — is the most consequential one-line calculation in twentieth-century physics. It convinced the founders of quantum electrodynamics that the infinities of perturbation theory could be tamed (renormalisation) and that the resulting theory had real predictive power. Today the electron $g-2$ is the most precisely tested prediction in all of physical science, with theory and experiment agreeing to twelve decimal places.

Visual Beginner

ONE-LOOP QUANTUM-ELECTRODYNAMICS VERTEX CORRECTION
================================

       e-(p)                              e-(p')
        \                                  /
         \                                /
          \                              /
           \----- (virtual gamma) ------/
                \                    /
                 \                  /
                  \                /
                   \  ___/\/\/\__/      <-- external photon (q = p' - p)
                    \/    ext     \/
                    /      gamma   \
                   /                \

   F_1(q^2) : Dirac form factor                    F_1(0) = 1     (Ward identity)
   F_2(q^2) : Pauli (anomalous-moment) form factor F_2(0) = alpha/(2 pi)
                                                          ~ 0.00116141

   g/2 = F_1(0) + F_2(0) = 1 + alpha/(2 pi)
   a_e = (g - 2)/2 = alpha/(2 pi) = 0.001161409...     (Schwinger 1948)

HISTORY AT A GLANCE
====================
                            measured g/2       comments
                            -------------      --------------------
   Dirac 1928 (tree):       1.000000           non-rel limit of Dirac eq.
   Foley-Kusch 1948:        1.00119(5)         first observed anomaly
   Schwinger 1948:          1.001161...        +alpha/(2 pi), one loop
   Petermann/Sommerfield:   +O(alpha/pi)^2     1957, two loops, analytic
   Laporta-Remiddi 1996:    +O(alpha/pi)^3     three loops, analytic
   Aoyama et al. 2018:      +O(alpha/pi)^5     numerical, five loops
   Hanneke-Fogwell-         1.00115965218073(28)  Penning trap, 2008
       Gabrielse 2008                           agreement at 10^{-12}

Object	Symbol	Role
Tree vertex	$-ie{\gamma }^{\mu }$	one electron-photon coupling, no loop
Dressed vertex	$-ie{\Gamma }^{\mu }(p^{\prime },p)$	tree + loops
Dirac form factor	$F_1(q^2)$	coefficient of ${\gamma }^{\mu }$; renormalised charge
Pauli form factor	$F_2(q^2)$	coefficient of $i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$; anomalous moment
Anomalous moment	$a_e=(g-2)/2=F_2(0)$	what Schwinger computed
Fine-structure constant	$\alpha \approx 1/137$	small parameter of the perturbation series

Worked example Beginner

Problem. Use the Schwinger one-line result $a_e=\alpha /(2\pi )$ to predict the anomalous magnetic moment of the electron. Compare with the most precise Penning-trap measurement $a_e^{\text{exp}}=0.00115965218073(28)$ (Hanneke, Fogwell, Gabrielse, 2008). Estimate the size of the two-loop correction.

Solution.

Step 1. Plug in $\alpha =1/137.035999084$ (CODATA 2018):

$$a_e^{(1)}=\frac{\alpha }{2\pi }=\frac{1}{2\pi \cdot 137.035999}=\frac{1}{861.022604}=0.001161409734\dots$$

Step 2. Subtract from the measured value:

$$a_e^{\text{exp}}-a_e^{(1)}=0.001159652181-0.001161409734=-0.000001757553$$

The one-loop prediction overshoots experiment by about $1.76\times 10^{-6}$, which is $1.5\times 10^{-3}$ of the one-loop value itself.

Step 3. Estimate the two-loop correction. Each additional loop costs a factor of $\alpha /\pi \approx 1/431$. So a generic two-loop coefficient of $O(1)$ would shift $a_e$ by $\Delta a_e\sim (\alpha /\pi {)}^2\approx 5.4\times 10^{-6}$. The actual two-loop coefficient, computed by Petermann and Sommerfield in 1957, is $-0.3284789\dots$, giving $a_e^{(2)}=-0.328\cdot (\alpha /\pi {)}^2=-1.77\times 10^{-6}$.

Step 4. Add the two-loop piece:

$$a_e^{(1)}+a_e^{(2)}=0.001161410-0.000001772=0.001159638$$

This already agrees with the measurement to six decimal places. The three-loop (Laporta-Remiddi 1996), four-loop, and five-loop (Aoyama et al. 2018) corrections close the remaining gap. The point of the worked example is that one ordinary-looking line of calculation — adding a single virtual photon to the bare vertex — captures essentially all of the electron's measured magnetic anomaly. The deeper layers of perturbation theory only correct it in the seventh decimal place.

Check your understanding Beginner

Formal definition Intermediate+

Setup. Consider the QED Lagrangian

$${\mathcal{L}}_{\text{QED}}=\bar{\psi }(i{\gamma }^{\mu }{\partial }_{\mu }-m)\psi -\frac{1}{4}{F}_{\mu \nu }{F}^{\mu \nu }-e\bar{\psi }{\gamma }^{\mu }\psi A_{\mu },$$

with the Feynman rules established in 12.12.01. Let $u(p)$ and $\bar{u}(p^{\prime })$ be on-shell Dirac spinors for incoming and outgoing electrons of four-momenta $p,p^{\prime }$ satisfying $\text{slashed}pu(p)=mu(p)$, $\bar{u}(p^{\prime })\text{slashed}{p}^{\prime }=m\bar{u}(p^{\prime })$, and let $q=p^{\prime }-p$ be the four-momentum transferred by the external photon (so $q^2\le 0$ for spacelike scattering kinematics).

The amputated electron-photon vertex function ${\Gamma }^{\mu }(p^{\prime },p)$ is defined as the sum of all one-particle-irreducible (1PI) Feynman diagrams with two amputated electron external lines and one amputated photon external line. To all orders in perturbation theory ${\Gamma }^{\mu }$ is a $4\times 4$ Dirac matrix-valued function of $p,p^{\prime }$. At lowest (tree) order ${\Gamma }^{\mu }={\gamma }^{\mu }$; the next contribution is the one-loop vertex diagram in which a virtual photon spans the two electron legs.

Form factor decomposition. Lorentz covariance, parity, charge-conjugation invariance, and the on-shell Dirac equation restrict the most general matrix element to exactly two independent Lorentz structures. Citing Berestetskii-Lifshitz-Pitaevskii §116 and Peskin-Schroeder §6.2 ^{[Berestetskii-Lifshitz-Pitaevskii §116; Peskin-Schroeder §6.2]}, one obtains the Gordon-decomposed vertex:

$$\boxed{\bar{u}(p^{\prime }){\Gamma }^{\mu }(p^{\prime },p)u(p)=\bar{u}(p^{\prime })\left[F_1(q^2){\gamma }^{\mu }+\frac{i{F}_2(q^2)}{2m}{\sigma }^{\mu \nu }{q}_{\nu }\right]u(p),}$$

where ${\sigma }^{\mu \nu }=\frac{i}{2}[{\gamma }^{\mu },{\gamma }^{\nu }]$ and the two real Lorentz-scalar functions

• $F_1(q^2)$ is the Dirac form factor (electric form factor, charge form factor), and
• $F_2(q^2)$ is the Pauli form factor (anomalous magnetic form factor),

are the on-shell observables that contain everything the photon can learn about a single dressed electron at one-photon exchange.

The Gordon identity

$$\bar{u}(p^{\prime }){\gamma }^{\mu }u(p)=\bar{u}(p^{\prime })\left[\frac{(p+p^{\prime }{)}^{\mu }}{2m}+\frac{i{\sigma }^{\mu \nu }{q}_{\nu }}{2m}\right]u(p)$$

is what makes this parameterisation natural: it isolates the convective current $(p+p^{\prime }{)}^{\mu }/(2m)$ from the spin current $i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$, and the Pauli structure ${\sigma }^{\mu \nu }{q}_{\nu }$ is the unique one that couples directly to spin via the Foldy-Wouthuysen reduction.

Physical content. At $q^2=0$ the two form factors are the on-shell observables seen by a static external field:

• $F_1(0)=1$ exactly to all orders in perturbation theory. This is enforced by the Ward-Takahashi identity (gauge invariance plus current conservation) together with the on-shell renormalisation convention that the renormalised electric charge $e$ is what Coulomb's law measures at long distance. Any one-loop UV divergence in $F_1(q^2)$ is removed by the vertex counterterm $Z_1-1$, and $Z_1=Z_2$ (vertex equals electron-wavefunction renormalisation) is itself a Ward-identity consequence.
• $F_2(0)=a_e=(g-2)/2$. The total $g$-factor of the electron is $$\frac{g}{2}=F_1(0)+F_2(0)=1+a_e.$$ At tree level $F_2=0$ and $g=2$ (Dirac). The Pauli form factor is generated entirely by loops.

For non-zero $q^2$ the form factors encode the spatial structure of the dressed electron: their Fourier transforms in the Breit frame give a charge density and a magnetisation density of size $\sim 1/m_e$ (a single Compton wavelength), with the vacuum polarisation of [12.16.03 — pending] adding the Uehling tail at the Coulomb potential.

Sign and metric conventions

Throughout this unit the metric is the mostly-plus particle-physics convention ${\eta }_{\mu \nu }=\mathrm{diag}(+1,-1,-1,-1)$, the covariant derivative is $D_{\mu }={\partial }_{\mu }+ie{A}_{\mu }$ (so the electron has charge $-e$ with $e>0$), the gamma matrices satisfy $\{{\gamma }^{\mu },{\gamma }^{\nu }\}=2{\eta }^{\mu \nu }1$, and on-shell spinors are normalised $\bar{u}(p)u(p)=2m$, ${\sum }_s{u}^s(p){\bar{u}}^s(p)=\text{slashed}p+m$, agreeing with 12.11.01. Berestetskii-Lifshitz-Pitaevskii use the opposite sign convention for $e$; readers consulting BLP should flip $e\to -e$ in the vertex factor.

Key derivation Intermediate+

Theorem (Schwinger 1948). The one-loop electron-photon vertex of QED gives

$$F_2(0)=\frac{\alpha }{2\pi }\simeq 1.161409\times 10^{-3}.$$

Consequently the anomalous magnetic moment of the electron at one loop is $a_e=\alpha /(2\pi )$.

Proof. Write the one-loop vertex correction $\delta {\Gamma }^{\mu }$ as the sum of the single 1PI diagram in which an electron of momentum $p$ emits a virtual photon of momentum $k$, scatters off the external photon $q$, and reabsorbs the virtual photon to leave as an electron of momentum $p^{\prime }$. The Feynman rules of QED (Feynman gauge) give

$$\bar{u}(p^{\prime })\delta {\Gamma }^{\mu }u(p)=(-ie{)}^2\int \frac{d^{4}k}{(2\pi {)}^4}\bar{u}(p^{\prime }){\gamma }^{\nu }\frac{i(\text{slashed}{p}^{\prime }+\text{slashed}k+m)}{(p^{\prime }+k{)}^2-m^2+iϵ}{\gamma }^{\mu }\frac{i(\text{slashed}p+\text{slashed}k+m)}{(p+k{)}^2-m^2+iϵ}{\gamma }_{\nu }\frac{-i}{k^2+iϵ}u(p).$$

Combine the three denominators using the Feynman-parameter identity

$$\frac{1}{A_1{A}_2{A}_3}=2{\int }_0^{1}d{x}_{1}d{x}_{2}d{x}_3\frac{\delta (1-x_1-x_2-x_3)}{(x_1{A}_1+x_2{A}_2+x_3{A}_3{)}^3}.$$

With $A_1=(p+k{)}^2-m^2$ (internal electron, propagating with momentum $p+k$), $A_2=(p^{\prime }+k{)}^2-m^2$, and $A_3=k^2$, the denominator becomes

$$x_1[(p+k{)}^2-m^2]+x_2[(p^{\prime }+k{)}^2-m^2]+x_3{k}^2={\ell }^2-\Delta +iϵ,$$

where the shifted loop momentum is $\ell =k+x_{1}p+x_2{p}^{\prime }$ and

$$\Delta =-x_1{x}_2{q}^2+(1-x_3{)}^2{m}^2=-x_1{x}_2{q}^2+(x_1+x_2{)}^2{m}^2.$$

The on-shell relations $p^2=p^{\prime 2}=m^2$ and $2p\cdot p^{\prime }=2m^2-q^2$ were used to land at this form; the $iϵ$ keeps Wick rotation legal.

The numerator, before the loop-momentum shift, is

$$N^{\mu }=\bar{u}(p^{\prime }){\gamma }^{\nu }(\text{slashed}{p}^{\prime }+\text{slashed}k+m){\gamma }^{\mu }(\text{slashed}p+\text{slashed}k+m){\gamma }_{\nu }u(p).$$

Shift $k\to \ell -x_{1}p-x_2{p}^{\prime }$ and use the standard gamma-matrix contraction identities

$${\gamma }^{\nu }{\gamma }^{\alpha }{\gamma }_{\nu }=-2{\gamma }^{\alpha },{\gamma }^{\nu }{\gamma }^{\alpha }{\gamma }^{\beta }{\gamma }_{\nu }=4{\eta }^{\alpha \beta }1,{\gamma }^{\nu }{\gamma }^{\alpha }{\gamma }^{\beta }{\gamma }^{\gamma }{\gamma }_{\nu }=-2{\gamma }^{\gamma }{\gamma }^{\beta }{\gamma }^{\alpha }.$$

The integrand splits into pieces even and odd under $\ell \to -\ell$; the odd pieces integrate to zero, and the surviving even-${\ell }^2$ piece is symmetrised by ${\ell }^{\mu }{\ell }^{\nu }\to \frac{1}{4}{\ell }^2{\eta }^{\mu \nu }$ (legitimate inside a rotationally-symmetric Euclidean integral). Apply the on-shell Dirac equations $\text{slashed}pu(p)=mu(p)$ and $\bar{u}(p^{\prime })\text{slashed}{p}^{\prime }=m\bar{u}(p^{\prime })$ wherever a $\text{slashed}p$ acts on $u(p)$ (or $\text{slashed}{p}^{\prime }$ on $\bar{u}(p^{\prime })$) to eliminate those terms in favour of $m$. After this algebra the numerator regroups into the Gordon-decomposed form

$$N^{\mu }=\bar{u}(p^{\prime })[{\gamma }^{\mu }\mathcal{A}({\ell }^2,x_i,q^2)+\frac{i{\sigma }^{\mu \nu }{q}_{\nu }}{2m}\mathcal{B}({\ell }^2,x_i,q^2)+(\text{anomaly-free convective term, recombined via Gordon})]u(p),$$

with two scalar coefficients $\mathcal{A},\mathcal{B}$. The coefficient $\mathcal{A}$ — the prospective contribution to $F_1$ — contains the ${\ell }^2$-piece, which after Wick rotation ${\ell }^0=i{\ell }_E^0$ and $d^4\ell \to i{d}^4{\ell }_E$ produces a logarithmically UV-divergent integral. The coefficient $\mathcal{B}$ — the prospective contribution to $F_2$ — has no ${\ell }^2$ in the numerator and is therefore UV-finite.

Specifically, after the Wick rotation and the angular integral, the contribution to $F_2$ is the convergent integral

$$F_2(q^2)=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_{1}d{x}_{2}d{x}_3\delta (1-x_1-x_2-x_3)\frac{2m^2{x}_3(1-x_3)}{m^2(1-x_3{)}^2-x_1{x}_2{q}^2}.$$

Set $q^2=0$ to extract the anomalous magnetic moment:

$$F_2(0)=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_{1}d{x}_{2}d{x}_3\delta (1-x_1-x_2-x_3)\frac{2x_3(1-x_3)}{(1-x_3{)}^2}=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_{1}d{x}_{2}d{x}_3\delta (1-x_1-x_2-x_3)\frac{2x_3}{1-x_3}.$$

Integrate out $x_2=1-x_3-x_1$ against the delta, with $0\le x_1\le 1-x_3$:

$$F_2(0)=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_3\frac{2x_3}{1-x_3}{\int }_0^{1-x_3}d{x}_1=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_3\frac{2x_3}{1-x_3}(1-x_3)=\frac{\alpha }{2\pi }{\int }_0^{1}d{x}_{3}2x_3.$$

The last integral is ${\int }_0^{1}2x_{3}d{x}_3=1$. Therefore

$$F_2(0)=\frac{\alpha }{2\pi }.\square$$

Everything that was UV-divergent ended up in $F_1$ and is removed by the vertex counterterm, with $F_1(0)=1$ enforced order-by-order by the Ward-Takahashi identity. Everything that was potentially IR-divergent — the $1/(1-x_3)$ singularity as $x_3\to 1$ — was cancelled by the explicit $1-x_3$ that the on-shell projection produced in the numerator. The Schwinger result $F_2(0)=\alpha /(2\pi )$ is finite, gauge-invariant, and IR-safe on its own.

Counterexamples to common slips

• The Schwinger calculation is not IR-finite at finite $q^2$ when external lines are off-shell; the IR-finiteness here is special to $F_2(0)$. Compare with the Sudakov double-logarithms in $F_1(q^2)$ at large $|q^2|$.
• $F_2(0)$ does not depend on the choice of photon gauge (Feynman vs. Lorenz vs. Coulomb vs. background-field) because $F_2$ is the coefficient of an on-shell projector. Working in Feynman gauge is a calculational convenience, not a physical assumption.
• The form factors $F_1,F_2$ are real for spacelike $q^2<4m^2$. The threshold at $q^2=4m^2$ corresponds to the electron-positron pair-production cut and gives imaginary parts at higher loop orders. The Schwinger result lives strictly below the cut.
• The Pauli term $i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$ is sometimes written as $-i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$ in older references; the sign tracks the convention for ${\sigma }^{\mu \nu }=\frac{i}{2}[{\gamma }^{\mu },{\gamma }^{\nu }]$ versus ${\sigma }^{\mu \nu }=-\frac{i}{2}[{\gamma }^{\mu },{\gamma }^{\nu }]$ — Berestetskii-Lifshitz-Pitaevskii use one sign, Peskin-Schroeder the other.

Exercises Intermediate+

Advanced results Master

The Schwinger calculation seeded a seventy-five-year programme of progressively higher-order QED corrections, and the experimental side kept pace through Penning-trap techniques that have improved by roughly an order of magnitude per decade since the 1980s.

Two-loop QED: Petermann 1957, Sommerfield 1957. A. Petermann (Helv. Phys. Acta 30, 407, 1957) and C. M. Sommerfield (Phys. Rev. 107, 328, 1957; Ann. Phys. (NY) 5, 26, 1958) independently obtained the closed-form two-loop result

$$a_e^{(2)}=\left(\frac{197}{144}+\frac{{\pi }^2}{12}-\frac{{\pi }^2}{2}\log 2+\frac{3}{4}\zeta (3)\right){\left(\frac{\alpha }{\pi }\right)}^2=-0.328478965\dots {\left(\frac{\alpha }{\pi }\right)}^2.$$

Seven two-loop topologies contribute; the appearance of $\zeta (3)$, ${\pi }^2\log 2$, and the polylogarithms anticipates the multiple-polylogarithm number-theoretic structure that recurs at every higher order. Two-loop vacuum-polarisation insertions (an electron loop inside the photon propagator that dresses the one-loop vertex) contribute the leading lepton-mass-dependent piece $\sim \log (m_{\ell }/m_e)$ that breaks the one-loop universality between species.

Three-loop QED: Laporta-Remiddi 1996. S. Laporta and E. Remiddi (Phys. Lett. B 379, 283, 1996) completed the three-loop analytic result

$$a_e^{(3)}=1.181241456\dots {\left(\frac{\alpha }{\pi }\right)}^3,$$

after a decades-long Bologna-Frascati effort building on Kinoshita and Cvitanovic. The analytic expression involves polylogarithms up to weight five, including the constants $\zeta (5)$, $\zeta (3{)}^2$, and Clausen functions. The systematisation of the loop-by-loop reduction to master integrals (Laporta algorithm) created the integration-by-parts technology that made all subsequent multi-loop calculations possible.

Four- and five-loop QED: Kinoshita-Aoyama-Hayakawa-Nio. T. Kinoshita and collaborators carried out a multi-decade computer-algebra programme that produced numerical values for the four-loop coefficient (891 diagrams) and ultimately the five-loop coefficient (12 672 diagrams). The Aoyama-Hayakawa-Kinoshita-Nio 2018 result (Phys. Rev. D 97, 036001) gives

$$a_e^{(5)}=6.737(159){\left(\frac{\alpha }{\pi }\right)}^5,$$

with the uncertainty arising from numerical Monte Carlo integration. The five-loop calculation required dedicated supercomputer time and an independent re-evaluation by Volkov 2019 (Phys. Rev. D 100, 096004) that found a 4% discrepancy in some subsets, presently under reconciliation.

Hadronic and electroweak contributions. Beyond pure QED, the electron $a_e$ receives a hadronic vacuum-polarisation contribution $a_e^{\text{had,VP}}\approx 1.6\times 10^{-12}$ and an electroweak contribution $a_e^{\text{EW}}\approx 0.03\times 10^{-12}$, both far below the experimental sensitivity. For the muon $a_{\mu }$ these contributions are amplified by $(m_{\mu }/m_e{)}^2\approx 4\times 10^4$ and become dominant uncertainties.

Experimental precision. Hanneke, Fogwell, and Gabrielse (Phys. Rev. Lett. 100, 120801, 2008; updated in Phys. Rev. A 83, 052122, 2011) measured a single electron in a cylindrical cryogenic Penning trap by detecting quantum jumps between cyclotron and spin levels, obtaining

$$a_e^{\text{exp}}=1.15965218073(28)\times 10^{-3},$$

a 0.24 ppt measurement. The theoretical Standard Model prediction (Aoyama et al. 2019, Atoms 7, 28, summarising the SM five-loop QED + hadronic + electroweak) is

$$a_e^{\text{SM}}=1.15965218161(23)\times 10^{-3},$$

agreeing with experiment within combined uncertainty. The result is now used in reverse: the most precise determination of $\alpha$ comes from inverting the five-loop QED calculation against the Penning-trap $a_e^{\text{exp}}$, giving ${\alpha }^{-1}=137.035999166(15)$ (Morel et al., Nature 588, 61, 2020, using a Rb-atom recoil determination as a cross-check). The agreement at the twelfth decimal place stands as the most stringent quantitative test of any theory in physics.

The muon anomaly. The muon's enhanced sensitivity to high-mass virtual particles, $\Delta a_{\mu }/\Delta a_e\sim (m_{\mu }/m_e{)}^2\approx 4\times 10^4$, makes it a precision probe of new physics. The CERN g-2 collaboration (1961-1979) measured $a_{\mu }$ to 7 ppm; the BNL E821 experiment (Bennett et al., Phys. Rev. D 73, 072003, 2006) reached 0.54 ppm; the Fermilab Muon g-2 experiment (Abi et al., Phys. Rev. Lett. 126, 141801, 2021; Aguillard et al., Phys. Rev. Lett. 131, 161802, 2023) currently quotes 0.20 ppm. The combined experimental average gives

$$a_{\mu }^{\text{exp}}=1.16592055(24)\times 10^{-3}.$$

The Standard Model prediction is contentious: using the $e^{+}{e}^{-}\to$ hadrons dispersion-relation determination of the leading hadronic vacuum-polarisation contribution (Aoyama et al., Phys. Rep. 887, 1, 2020; Theory Initiative white paper) gives $a_{\mu }^{\text{SM}}=1.16591810(43)\times 10^{-3}$, differing from experiment by $4.2\sigma$. Using the BMW 2020 lattice-QCD calculation of the same contribution (Borsanyi et al., Nature 593, 51, 2021), the SM prediction shifts upward and the discrepancy with experiment shrinks to about $1.5\sigma$. Resolving this tension between dispersion-relation and lattice determinations of the hadronic vacuum polarisation is currently the central open problem in precision QED tests; the answer determines whether the muon $g-2$ is the first quantitative laboratory hint of physics beyond the Standard Model.

Beyond-QED probes. A supersymmetric partner of mass $M$ contributes $\Delta a_{\ell }\sim (\alpha /4\pi )(m_{\ell }/M{)}^2\tan \beta$ to $a_{\ell }$, giving the muon $a_{\mu }$ an SUSY sensitivity that probes superpartner masses out to a few TeV. Leptoquarks, $Z^{\prime }$ bosons, two-Higgs-doublet extensions, and dark photons all contribute analogously, and the BNL/Fermilab muon $g-2$ result is now part of every BSM model's parameter-space fit. For the electron $a_e$, the recoil-based determination of $\alpha$ at the Berkeley (Parker et al., Science 360, 191, 2018, with Cs atoms) and the LKB / Müller (Morel et al., 2020, with Rb atoms) experiments gives discrepant values that translate to a $2.4\sigma$ tension in $a_e^{\text{SM}}-a_e^{\text{exp}}$, again left as a current open problem with implications for light-dark-matter and dark-photon searches.

Theory technology. The Schwinger-Tomonaga-Feynman-Dyson reconciliation in 1948-1949 (Dyson's two papers in Phys. Rev. 75, 486 and 75, 1736, 1949) established the covariant perturbation theory, the renormalisation programme, and the Feynman diagram graphical calculus all in a year. The same year, Furry's theorem (W. H. Furry, Phys. Rev. 51, 125, 1937) cleared the way for the calculation by ruling out closed fermion-loop diagrams with an odd number of photon attachments — without Furry, the one-loop vertex would have had a competing $C$-violating diagram and the universality of $F_2(0)$ would be obscured. The Ward-Takahashi identity (J. C. Ward, Phys. Rev. 78, 182, 1950; Y. Takahashi, Nuovo Cim. 6, 371, 1957) made $F_1(0)=1$ a consequence of gauge invariance rather than an accident of the one-loop calculation, and that elevation is what survived the subsequent generalisation to non-Abelian gauge theory by 't Hooft and Veltman in 1971-1972.

Full proof set Master

The Key derivation supplies the on-shell $q^2=0$ evaluation in full. The following auxiliary results are stated with proof outlines that the reader can verify against the cited literature.

Lemma 1 (Gordon decomposition uniqueness). Lorentz covariance, parity, $C$-conjugation, the on-shell Dirac equation, and the $4\times 4$ Dirac algebra reduce the most general one-photon-irreducible matrix element $\bar{u}(p^{\prime }){\Gamma }^{\mu }u(p)$ to exactly two independent structures: ${\gamma }^{\mu }$ and $i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$. Proof outline. The general matrix element is a linear combination of $\{{\gamma }^{\mu },p^{\mu },p^{\prime \mu },{\gamma }^{\mu }\text{slashed}p,{\gamma }^{\mu }\text{slashed}{p}^{\prime },p^{\mu }\text{slashed}{p}^{\prime },\dots \}$. Using $\text{slashed}pu(p)=mu(p)$ and $\bar{u}(p^{\prime })\text{slashed}{p}^{\prime }=m\bar{u}(p^{\prime })$ eliminates all $\text{slashed}p$-on-right and $\text{slashed}{p}^{\prime }$-on-left structures. The Gordon identity converts ${\gamma }^{\mu }\to (p+p^{\prime }{)}^{\mu }/(2m)+i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$, and the $(p+p^{\prime }{)}^{\mu }$ structure is equivalent (on-shell) to ${\gamma }^{\mu }-i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$, leaving the two-term basis $\{{\gamma }^{\mu },i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)\}$. Parity and $C$ forbid the ${\gamma }^5$-projected structures ${\gamma }^{\mu }{\gamma }^5$ and ${\sigma }^{\mu \nu }{q}_{\nu }{\gamma }^5$ for an unbroken-$P$, unbroken-$C$ theory like QED. $\square$ (Detailed proof: BLP §116, or Itzykson-Zuber Ch. 7.)

Lemma 2 (Ward-Takahashi for the on-shell vertex). Current conservation ${\partial }_{\mu }{j}^{\mu }=0$ at the operator level forces $q_{\mu }{\Gamma }^{\mu }(p^{\prime },p)=S^{-1}(p^{\prime })-S^{-1}(p)$, with $S(p)$ the full electron propagator. Proof outline. Translate $⟨\Omega |T{j}^{\mu }(x)\psi (y)\bar{\psi }(z)|\Omega ⟩$ by ${\partial }_{\mu }$, use ${\partial }_{\mu }{j}^{\mu }=0$ on the operator, and pick up only the equal-time canonical commutators $[j^0,\psi ]$ and $[j^0,\bar{\psi }]$. The two delta-function terms produce $S(p^{\prime })-S(p)$ in momentum space, and amputating gives the stated identity. (Detailed proof: Peskin-Schroeder §7.4; Weinberg Vol. I §10.4.) $\square$

Lemma 3 (UV finiteness of $F_2$). The Pauli form factor at one loop is UV-finite. Proof outline. Power counting: the loop integral has $d^4\ell$ in the measure, three propagators contributing ${\ell }^{-6}$ at large $\ell$, and the numerator contributes at most ${\ell }^2$ (from two propagator numerators). The naive degree of divergence is $4-6+2=0$ (logarithmic). The leading log lives in the ${\gamma }^{\mu }$ coefficient (the ${\ell }^2$-piece, hence $F_1$), and the $i{\sigma }^{\mu \nu }{q}_{\nu }/(2m)$ coefficient picks up only ${\ell }^0$ terms in the numerator, reducing its degree to $-2$ (UV-finite). (Detailed proof: Peskin-Schroeder §6.3 displays the integral and verifies term-by-term.) $\square$

Lemma 4 (IR finiteness of $F_2(0)$). The Pauli form factor at $q^2=0$ is IR-finite. Proof outline. The IR divergence of the one-loop QED vertex is associated with the $k\to 0$ region of the virtual photon. In this region the integrand has the form $1/(k^2\cdot k\cdot p\cdot k\cdot p^{\prime })$ which is logarithmically divergent — but only when contracted with ${\gamma }^{\mu }$ via the eikonal numerator structure. The Pauli form factor's ${\sigma }^{\mu \nu }{q}_{\nu }$ structure requires a single power of $q$ from the loop, and at $q=0$ the eikonal coupling vanishes, leaving the residual integral IR-safe. (Detailed proof: BLP §117, with explicit IR-finiteness check.) $\square$

Lemma 5 (universality of the one-loop $a_{\ell }$ in pure QED). The lepton-mass-independence of $a_{\ell }^{(1)}=\alpha /(2\pi )$ follows from the dimensional analysis of the on-shell Feynman-parameter integral. Proof outline. The integrand at $q^2=0$ depends on $m$ only through the on-shell projector $(1-x_3{)}^2{m}^2$ in $\Delta$. The numerator on-shell projection produces a compensating $(1-x_3)$, and the Feynman-parameter integration over $x_1,x_2$ at fixed $x_3$ produces another $(1-x_3)$ from the triangular region $\{(x_1,x_2):x_1+x_2=1-x_3,x_1,x_2\ge 0\}$. Both factors of $(1-x_3)$ exactly cancel the $(1-x_3{)}^2$ in the denominator, leaving a mass-independent result. (Cf. BLP §117 and the worked Feynman-parameter integral above.) $\square$

Connections Master

• 12.11.01 supplies Dirac's tree-level $g=2$ from the non-relativistic Foldy-Wouthuysen reduction; this unit perturbs that bare value by adding a single virtual photon.

• 12.12.01 supplies the Feynman rules of QED (vertex factor, electron and photon propagators, on-shell spinor normalisation) that this unit composes into the vertex correction.

• [12.16.01 — pending] Electron self-energy and mass renormalisation dresses the propagator $S(p)$ by the same one-loop photon, and the Ward identity $Z_1=Z_2$ links its on-shell residue to the vertex renormalisation that fixes $F_1(0)=1$ in this unit.

• [12.16.03 — pending] Vacuum polarisation and the Uehling correction to Coulomb scattering is the photon-side companion: a virtual electron-positron loop dresses the photon propagator and shifts the Coulomb potential at short distance, contributing to the Lamb shift at the next order.

• [12.16.04 — pending] Lamb shift in the Bethe-Welton-Feynman style combines the self-energy, vertex, and vacuum-polarisation results into the hydrogen $2S_{1/2}$-$2P_{1/2}$ splitting, the second great quantitative success of one-loop quantum electrodynamics.

• [12.16.05 — pending] Bloch-Nordsieck and the cancellation of infrared divergences explains why the IR divergence in $F_1(q^2)$ at $q^2\ne 0$ cancels against soft-photon bremsstrahlung in any inclusive cross-section.

• [12.16.06 — pending] Two- and higher-loop $a_{\ell }$: Petermann-Sommerfield and Laporta-Remiddi upgrades this unit's one-loop result to the modern five-loop SM prediction and explains the Laporta integration-by-parts machinery.

• 12.05.05 Free Dirac quantum field gives the canonical-quantisation framework — CAR, mode expansion, Feynman propagator — that the vertex integral uses.

• 03.09.02 Clifford algebra underlies the gamma-matrix identities $\{{\gamma }^{\mu },{\gamma }^{\nu }\}=2{\eta }^{\mu \nu }$ on which every contraction in the Key derivation depends.

• 12.06.04 Crossing symmetry and the CPT theorem lies one step downstream: the same on-shell vertex is what crosses between $e^{-}\gamma \to e^{-}$ and $e^{+}\gamma \to e^{+}$ at the diagrammatic level, with $F_2(0)$ flipping sign between particle and antiparticle.

Historical & philosophical context Master

The anomalous magnetic moment of the electron began as an experimental puzzle. Rabi's Columbia molecular-beam group had been refining the hyperfine spectroscopy of hydrogen and gallium through the 1930s, and by 1947 Foley and Kusch — using a new atomic-beam magnetic-resonance apparatus — could measure the ratio of the Lande $g$-factors of two atomic states to a precision of a few parts in $10^4$. They found a small but persistent excess over the Dirac prediction $g=2$. The result was announced at the Shelter Island conference in June 1947, in the same session in which Lamb presented his measurement of the $2S_{1/2}$-$2P_{1/2}$ hydrogen splitting (later named the Lamb shift). Both experiments demanded an explanation that exceeded the single-particle Dirac equation.

Schwinger's calculation, completed in late 1947 and submitted in December for the February 1948 issue of Physical Review (Schwinger, Phys. Rev. 73, 416, 1948), arrived first and arrived clean. He computed $\alpha /(2\pi )$ as the leading radiative correction to the electron's magnetic moment, in a one-page note that simultaneously established the renormalisation programme would work and gave a number that matched Foley-Kusch within their experimental error. Schwinger immediately followed with a three-paper series (Phys. Rev. 74, 1439, 1948; 75, 651, 1949; 76, 790, 1949) developing the covariant operator formalism, the action principle, and the systematic computation of one-loop radiative corrections. Tomonaga had developed an equivalent formalism in occupied Japan during the war (Prog. Theor. Phys. 1, 27, 1946; translated and published in Phys. Rev. 74, 224, 1948) and reached the same result for $a_e$ independently. Feynman's space-time approach, presented at the 1948 Pocono conference and published as Phys. Rev. 76, 769, 1949, gave the calculation in its now-canonical Feynman-diagram language. Dyson's 1949 papers (Phys. Rev. 75, 486 and 75, 1736) proved that the three formalisms — Tomonaga's, Schwinger's, and Feynman's — were equivalent, and that the renormalisation programme worked order-by-order in perturbation theory.

The Nobel committee recognised the achievement in 1965, awarding the prize jointly to Schwinger, Tomonaga, and Feynman for "their fundamental work in quantum electrodynamics, with deep-ploughing consequences for the physics of elementary particles." Dyson, who had supplied the proof of internal consistency that the other three had presupposed, was famously not included; the customary three-recipient limit forced the choice. Foley and Kusch shared in the 1955 prize that went to Lamb (for the Lamb shift) and Polykarp Kusch (for the electron magnetic-moment determination), recognising the experimental discoveries that had triggered the theoretical work.

The two-loop calculation by Petermann and Sommerfield in 1957 (Helv. Phys. Acta 30, 407; Phys. Rev. 107, 328) proved that the Schwinger leading-order result was not a coincidence: the same machinery extended consistently to higher orders, with calculable coefficients that matched ever-more-precise measurements. Petermann's calculation was done by hand and contained an algebraic error that Sommerfield independently caught; the corrected value $a_e^{(2)}=-0.328478\cdot (\alpha /\pi {)}^2$ has stood since. The three-loop result of Laporta and Remiddi (Phys. Lett. B 379, 283, 1996) closed an open problem from the 1980s and required the development of computer-algebra techniques (FORM, the Laporta algorithm for integration-by-parts reductions) that subsequently revolutionised every other multi-loop calculation in particle physics. The four- and five-loop Aoyama-Hayakawa-Kinoshita-Nio results (Phys. Rev. Lett. 109, 111808, 2012; Phys. Rev. D 97, 036001, 2018) extend the precision QED calculation to twelve significant figures, matching the experimental precision of the Hanneke-Fogwell-Gabrielse 2008 Penning-trap measurement.

The muon $g-2$ programme — the CERN measurements of the 1960s and 1970s, the BNL E821 result of 2006, the Fermilab Muon g-2 results of 2021 and 2023 — has run in parallel and currently shows a roughly $4\sigma$ tension with the dispersion-relation-based Standard Model prediction, or a $1.5\sigma$ tension with the lattice-QCD-based prediction. Resolving this tension between two independent determinations of the leading hadronic vacuum-polarisation contribution is the central open question in precision QED tests as of 2026. If the lattice value is correct, the muon $g-2$ is consistent with the Standard Model and gives no hint of new physics; if the dispersion-relation value is correct, it is the first quantitative laboratory anomaly suggesting BSM physics at the TeV scale.

Bibliography Master

Primary literature:

• Foley, H. M. & Kusch, P. On the intrinsic moment of the electron. Phys. Rev. 73, 412 (1948); Phys. Rev. 74, 250 (1948). The experimental discovery that $g\ne 2$.

• Schwinger, J. On quantum-electrodynamics and the magnetic moment of the electron. Phys. Rev. 73, 416 (1948). The original one-page derivation of $a_e^{(1)}=\alpha /(2\pi )$.

• Schwinger, J. Quantum electrodynamics. I, II, III. Phys. Rev. 74, 1439 (1948); 75, 651 (1949); 76, 790 (1949). The systematic covariant operator-formalism development.

• Tomonaga, S. On a relativistically invariant formulation of the quantum theory of wave fields. Prog. Theor. Phys. 1, 27 (1946); English version Phys. Rev. 74, 224 (1948). The independent Japanese-wartime derivation.

• Feynman, R. P. Space-time approach to quantum electrodynamics. Phys. Rev. 76, 769 (1949). The diagrammatic formulation of the calculation.

• Dyson, F. J. The radiation theories of Tomonaga, Schwinger, and Feynman. Phys. Rev. 75, 486 (1949); The S-matrix in quantum electrodynamics. Phys. Rev. 75, 1736 (1949). The proof of equivalence and renormalisability.

• Petermann, A. Fourth order magnetic moment of the electron. Helv. Phys. Acta 30, 407 (1957). Two-loop analytic result.

• Sommerfield, C. M. Magnetic dipole moment of the electron. Phys. Rev. 107, 328 (1957); Ann. Phys. (NY) 5, 26 (1958). Independent two-loop confirmation.

• Laporta, S. & Remiddi, E. The analytical value of the electron (g-2) at order alpha^3 in quantum electrodynamics. Phys. Lett. B 379, 283 (1996). Three-loop analytic result.

• Hanneke, D., Fogwell, S. & Gabrielse, G. New measurement of the electron magnetic moment and the fine structure constant. Phys. Rev. Lett. 100, 120801 (2008); Phys. Rev. A 83, 052122 (2011). The 0.24 ppt Penning-trap measurement.

• Aoyama, T., Hayakawa, M., Kinoshita, T. & Nio, M. Complete tenth-order quantum-electrodynamics contribution to the muon g-2. Phys. Rev. Lett. 109, 111808 (2012); Tenth-order electron anomalous magnetic moment: Contribution of diagrams without closed lepton loops. Phys. Rev. D 91, 033006 (2015); Theory of the anomalous magnetic moment of the electron. Atoms 7, 28 (2019). The five-loop programme.

• Bennett, G. W. et al. (Muon g-2 Collaboration). Final report of the muon E821 anomalous magnetic moment measurement at BNL. Phys. Rev. D 73, 072003 (2006). The BNL muon $a_{\mu }$ result.

• Abi, B. et al. (Muon g-2 Collaboration). Measurement of the positive muon anomalous magnetic moment to 0.46 ppm. Phys. Rev. Lett. 126, 141801 (2021); Aguillard, D. P. et al. Measurement of the positive muon anomalous magnetic moment to 0.20 ppm. Phys. Rev. Lett. 131, 161802 (2023). The Fermilab muon $a_{\mu }$ results.

• Aoyama, T. et al. The anomalous magnetic moment of the muon in the Standard Model. Phys. Rep. 887, 1 (2020). The Theory Initiative white paper consolidating the SM prediction.

• Borsanyi, S. et al. (BMW Collaboration). Leading hadronic contribution to the muon magnetic moment from lattice QCD. Nature 593, 51 (2021). The lattice-QCD hadronic vacuum-polarisation result currently in tension with dispersion-relation evaluations.

• Morel, L., Yao, Z., Clade, P. & Guellati-Khelifa, S. Determination of the fine-structure constant with an accuracy of 81 parts per trillion. Nature 588, 61 (2020). The recoil-based atomic determination of $\alpha$.

• Parker, R. H., Yu, C., Zhong, W., Estey, B. & Müller, H. Measurement of the fine-structure constant as a test of the Standard Model. Science 360, 191 (2018). The Berkeley Cs-recoil $\alpha$ determination.

Textbook treatments:

• Berestetskii, V. B., Lifshitz, E. M. & Pitaevskii, L. P. Quantum Electrodynamics. Vol. 4 of the Landau-Lifshitz Course of Theoretical Physics, 2e. Pergamon / Butterworth-Heinemann, 1982. §116 (vertex function general framework) and §117 (Schwinger anomalous moment). The canonical physicist-process-driven treatment.

• Weinberg, S. The Quantum Theory of Fields, Vol. I: Foundations. Cambridge University Press, 1995. §10 (one-loop renormalisation), §11.3 (anomalous magnetic moment), with a careful Ward-identity discussion.

• Peskin, M. E. & Schroeder, D. V. An Introduction to Quantum Field Theory. Westview Press, 1995. §6.2 (form factor decomposition), §6.3 (the Schwinger calculation in detail, with all Feynman-parameter manipulations explicit).

• Schwartz, M. D. Quantum Field Theory and the Standard Model. Cambridge University Press, 2014. Ch. 17 (electron magnetic moment) and Ch. 18 (mass renormalisation), updated through the four-loop result.

• Itzykson, C. & Zuber, J.-B. Quantum Field Theory. McGraw-Hill, 1980. Ch. 7 (radiative corrections), with explicit gamma-matrix manipulations.

• Bjorken, J. D. & Drell, S. D. Relativistic Quantum Mechanics (1964) and Relativistic Quantum Fields (1965). McGraw-Hill. The first systematic textbook treatment with the explicit vertex calculation.

• Jegerlehner, F. The Anomalous Magnetic Moment of the Muon. Springer Tracts in Modern Physics 274, 2017. Comprehensive monograph on $a_{\mu }$ theory and experiment.